CompliLearn

Turn dense compliance docs into voice-friendly, quizzed microlearning, then bring the same policy intelligence into your coding assistant with MCP.

Comment

Inspiration

Compliance training is often a one-time PDF dump: people skim, click through, and forget. We were inspired by a simple problem we kept seeing in onboarding: teams are expected to follow strict policies, but the learning experience is passive and disconnected from real work.

We wanted to make compliance feel more like a coach than a checklist.

What We Built

We built CompliLearn, an AI-powered compliance learning platform with two connected experiences:

Training experience for learners

  • Upload compliance documents (like HIPAA or internal policy PDFs)
  • Auto-generate bite-sized modules, plain-language summaries, and quizzes
  • Add voice playback for accessibility and better retention
  • Track module progress and quiz outcomes

Compliance context for coding assistants (MCP)

  • Expose policy-aware tools via a Model Context Protocol (MCP) server
  • Let AI coding agents query policy sections and run compliance checks while generating code
  • Keep policy guidance present during implementation, not just onboarding

How We Built It

  • Frontend: Next.js + React + TypeScript + Tailwind/shadcn UI
  • Backend: Convex (database, auth, file storage, server functions)
  • AI pipeline: Document ingestion + retrieval + generation workflows for summaries, quizzes, and Q&A
  • Voice: ElevenLabs integration for narrated module content
  • Developer integration: Standalone MCP server for policy tools used by editor agents

We split development across the team and worked in parallel:

  • UI Prototype with Figma Make
  • Training UX and accessibility
  • Backend ingestion/retrieval flows
  • MCP integration and demo reliability

Challenges We Ran Into

  • Turning long PDFs into useful learning content: Preserving policy accuracy while simplifying language.
  • Latency and orchestration: Chaining upload, processing, generation, and rendering without a confusing UX.
  • Hackathon integration constraints: Getting multiple moving parts (AI, voice, auth, MCP) stable enough for a live demo flow.
  • Grounding and trust: Making sure responses stay tied to policy source material instead of generic model answers.

What We Learned

  • Good compliance UX is mostly a product design problem, not just an AI problem.
  • Accessibility features (voice, chunked content) help everyone, not only edge cases.
  • MCP is a powerful bridge between "learning policy" and "applying policy in real work."
  • Shipping a coherent end-to-end experience in hackathon time requires ruthless scope discipline and clear interface boundaries.

What's Next

  • Replace all mock policy paths in MCP with fully live backend retrieval.
  • Add richer admin analytics for completion and weak-topic detection.
  • Improve citation transparency so every generated answer clearly maps to source sections.
  • Expand support for more policy frameworks and multilingual training.

Built With

Share this project:

Updates